Research On Prediction Model Of Heterosis Of Brassica Napus L

Brassica napus L as a the herb of the Brassica,belongs to the Cruciferae family.As the main oil and cash crops,the breeding of high-yielding Brassica napus L varieties by using heterosis is the breeding direction that breeders pay close attention to.In heterosis breeding,the number of hybrid combinations available for configuration far exceeds the number of actual selection by breeders.It is of great theoretical and practical significance to construct a simple and effective prediction index system to guide parent selection and to predict hybrid combinations.In this research,35 parents and 306 hybrid F1 were used as experimental materials.Using phenotypic and genotype data for 9 traits(Thousand seeds weight,Seeds per siliques,Seeds yield,Plant height,Effective length of main inflorescence,Branch number,Branch height,Siliques of main inflorescence,Siliques length)to explore the predictive ability of different predictive models.Prediction models achieve effective prediction of hybrid combinations of yield traits.The main results are as follows:1.Brassica napus L phenotypic genetic analysisStatistical and genetic analysis of the parental and hybrid F1 phenotype data showed that the phenotype data of each trait conformed to a normal distribution,and the phenotypic value variation coefficient ranged from 4.61% to 23.06%.The hybrid F1 showed obvious Mid-parent heterosis and Over-parent heterosis,and the narrow heritability and broad heritability of each trait respectively ranged from 0.156 to 0.667 and from 0.312 to 0.684.2.Genome-wide selection prediction modelsDifferent whole-genome selection models were used to analyze the 9 traits of hybrids,and the prediction accuracy of the models was evaluated by five-fold crossvalidation.The results showed that there was no significant difference in the prediction accuracy of different models for the same trait.There were significant differences in the prediction accuracy of among different traits.The prediction accuracy were highly correlated with the heritability of the traits,with a correlation coefficient of 0.988.The GBLUP model was used to explore the effect of marker density and population size on the prediction accuracy of genome-wide selection.The results showed that both marker density and population size had an impact on the prediction accuracy of GS.With the increase of marker density and population size,it first increased and then stabilized.For thousand seeds weight,seeds per siliques,seeds yield and siliques length,the number of markers required to achieve the best prediction effect is within 1000;for branch height and plant height,the number of markers required to achieve the best prediction effect is within 5000.The prediction accuracy of genome-wide selection for other traits reached the best prediction effect when the population number reached 250 except for branch number.3.The GBLUP model based on significant loci of GWASGWAS analysis was performed using 35 parents and 306 hybrids,and GBLUP model was established by screening significant association loci at 6 threshold levels.Compared with the GS model,its prediction accuracy has been improved to varying degrees.it indicated that the selection of loci associated with the trait can improve the prediction effect of the model.The markers screening different threshold levels of the optimal models for different traits were different,and the predicted correlation coefficients of the optimal models ranged from 0.391 to 0.737.The prediction accuracy was consistent with the GS result and was also affected by the heritability of the trait.4.GBLUP model based on between-group difference selection markersThe research materials were grouped by marker type as a fixed factor,and 15789 specific loci associated with 9 traits were screened at the 0.0001 significance level.Among them,the specific loci related to thousand seeds weight were the most,and the number of the loci is 9190.the specific locus related to branche numbers were the least,and the number of the loci is 155.According to the order of the absolute value of the heterosexual loci effect,the GBLUP model was established by setting 13 marker quantity levels.When the number of specific loci was about 300,the best prediction effect of the model could be achieved.Compared with the GBLUP model established by genome-wide markers,the GBLUP model established by each trait-specific locus has different degrees of improvement,and the improvement range is between 0.0005 and 0.1372.5.Multivariate linear model based on between-group difference selection markersAccording to the marker effect value of the specific locus and the phenotype value,a multivariate linear model was constructed by stepwise regression analysis.The coefficient of determination between the predicted value and the phenotype value of each trait ranged from 0.351 to 0.966.The model was tested with 30 repetitions of five-fold cross-validation,and the coefficient of determination of the prediction model was between 0.247 and 0.913.The number of markers involved in each prediction model was between 9 and 92,among which prediction model of thousand seeds weight had the largest number of markers,prediction model of branch number had the least number of markers.There were 11 markers common to the 9 traits.6.Predictive model evaluationUsing 5-fold cross-validation with 30 repetitions,various prediction models in the research were compared and analyzed.It was found that the correlation coefficient of the prediction model established using the molecular markers related to the trait was significantly higher than that of the genome-wide selection model.The linear model had the best prediction effect in the thousand seeds weight,seeds per siliques,seeds yield,plant height,effective length of main inflorescence,branch number,branch height and siliques of main inflorescence.The GBLUP model based on GWAS significant markers in silique length has the best prediction effect.